작성일: 2022.08.17 (Wed)
How to handle the case when the validation loss might go up and down? In that case, early stopping might prevent my model from learning further, right?
People typically define a patience, i.e. the number of epochs to wait before early stop if no progress on the validation set. The patience is often set comewhere between 10 and 100 (10 or 20 is more common), but it really depends on your dataset and network.
summary
validation loss의 변화가 없을 때, early stopping 동작함.
몇번의 epoch 이후, 변화 체크할 것인지를 나타내는 값인 patience는 지정가능함.
patience는 주로 10 ~ 100 사이로 정해지며 (10 또는 20이 가장 흔함), 해당 모델의 데이터셋과 네트워크에 따라 달라짐.
What is Early Stopping?
e.g. In case of monitor=val_loss
Reference
